Skip to content

Krug 2020

Verdict: The foundational CPTAC phosphoproteomics study that proved phospho-based subtypes add clinical value beyond genomics -- every subsequent CPTAC phospho paper builds on this template.

Citation: Krug K, Jaehnig EJ, Satpathy S, et al. "Proteogenomic Landscape of Breast Cancer Subtypes from the CPTAC Cohort." Cell 183(5):1436-1456.e31 (2020). DOI: 10.1016/j.cell.2020.10.036

Problem Setup

Breast cancer is classified by PAM50 transcriptomic subtypes (Luminal A, Luminal B, HER2-enriched, Basal-like). Can proteomics and phosphoproteomics reveal biology and clinical stratification that transcriptomics misses? Specifically, do phospho-based subtypes capture druggable signaling states invisible to gene expression?

Approach

Deep proteogenomic profiling of 122 treatment-naive breast cancers from the CPTAC cohort. TMT-based quantitative proteomics and phosphoproteomics (TiO2 enrichment) with matched whole-genome sequencing, RNA-seq, and clinical data. Phospho-based subtypes derived by consensus clustering on phosphosite abundance. Kinase activity inferred from substrate phosphorylation using kinase-substrate enrichment analysis (KSEA). Integration across omics layers to connect genomic drivers to phospho-signaling states.

Key Findings

  • Phospho-based clustering identifies subtypes that refine PAM50 classifications -- particularly splitting Luminal A/B into subgroups with distinct kinase activities and clinical outcomes.
  • ERBB2 amplification drives consistent phospho-signaling signatures, but the downstream targets vary between patients, suggesting combinatorial therapy opportunities.
  • Basal-like tumors show high CDK and DNA damage response kinase activity, consistent with PARP inhibitor sensitivity.
  • Proteomics-based immune subtypes correlate with but are not identical to transcriptomic immune signatures.
  • Acetylation data adds another regulatory layer, particularly for metabolic reprogramming in Luminal tumors.

Evaluation

Phospho subtypes validated for survival associations within the CPTAC cohort. Kinase activity predictions compared against known driver biology (e.g., ERBB2 amplification should elevate ERBB2 kinase activity). Cross-referencing with drug sensitivity databases (GDSC, CTRP) to propose therapeutic vulnerabilities. No prospective clinical validation or independent external cohort.

Honest Assessment

Strengths:

  • Established the template for CPTAC multi-omics analysis that all subsequent cancer type studies followed.
  • Demonstrated clearly that phosphoproteomics adds non-redundant information to transcriptomic subtypes.
  • High-quality, deeply characterized cohort with matched multi-omics at every level.
  • Data publicly available through the CPTAC data portal, enabling extensive reanalysis by the community.

Limitations:

  • 122 samples is small for subtype discovery -- statistical power for rare subtypes is limited.
  • TMT ratio compression reduces the dynamic range of phosphosite quantification, potentially attenuating real biological differences.
  • Bulk tumor profiling mixes tumor, stromal, and immune cell phospho-signals without deconvolution.
  • No functional validation of the proposed druggable kinase dependencies -- all therapeutic predictions remain computational.

Design Decision: Invest in depth of characterization (many omics layers, thorough integration) rather than cohort size. For a foundational study, this was the right choice -- it proved the concept that phosphoproteomics adds clinical value. Scaling to larger cohorts (as Geffen 2023 later did) required this proof-of-concept first.